[CostModel][X86] Add explicit int/fp arithmetic test coverage for avx512/slm/glm...

diff --git a/llvm/test/Analysis/CostModel/X86/arith-costkinds.ll b/llvm/test/Analysis/CostModel/X86/arith-costkinds.ll
index 85ea528..e981b19 100644 (file)
--- a/llvm/test/Analysis/CostModel/X86/arith-costkinds.ll
+++ b/llvm/test/Analysis/CostModel/X86/arith-costkinds.ll
@@ -18,6 +18,14 @@
 ; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512bw,+avx512vl -cost-kind=latency      < %s | FileCheck %s --check-prefixes=LATE
 ; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512bw,+avx512vl -cost-kind=code-size    < %s | FileCheck %s --check-prefixes=SIZE
 ; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512bw,+avx512vl -cost-kind=size-latency < %s | FileCheck %s --check-prefixes=SIZE_LATE
+;
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mcpu=slm -cost-kind=latency      < %s | FileCheck %s --check-prefixes=LATE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mcpu=slm -cost-kind=code-size    < %s | FileCheck %s --check-prefixes=SIZE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mcpu=slm -cost-kind=size-latency < %s | FileCheck %s --check-prefixes=SIZE_LATE
+;
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mcpu=goldmont -cost-kind=latency      < %s | FileCheck %s --check-prefixes=LATE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mcpu=goldmont -cost-kind=code-size    < %s | FileCheck %s --check-prefixes=SIZE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mcpu=goldmont -cost-kind=size-latency < %s | FileCheck %s --check-prefixes=SIZE_LATE
 
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
 target triple = "x86_64-apple-macosx10.8.0"

diff --git a/llvm/test/Analysis/CostModel/X86/arith-fp-costkinds.ll b/llvm/test/Analysis/CostModel/X86/arith-fp-costkinds.ll
index c8a45d8..760f762 100644 (file)
--- a/llvm/test/Analysis/CostModel/X86/arith-fp-costkinds.ll
+++ b/llvm/test/Analysis/CostModel/X86/arith-fp-costkinds.ll
@@ -1,15 +1,31 @@
 ; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse2 -cost-kind=latency      < %s | FileCheck %s --check-prefixes=CHECK,LATE,SSE-LATE
-; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse2 -cost-kind=code-size    < %s | FileCheck %s --check-prefixes=CHECK,SIZE,SSE-SIZE
-; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse2 -cost-kind=size-latency < %s | FileCheck %s --check-prefixes=CHECK,SIZE_LATE,SSE-SIZE_LATE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse2 -cost-kind=latency      < %s | FileCheck %s --check-prefixes=CHECK,LATE,SSE-LATE,SSE2-LATE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse2 -cost-kind=code-size    < %s | FileCheck %s --check-prefixes=CHECK,SIZE,SSE-SIZE,SSE2-SIZE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse2 -cost-kind=size-latency < %s | FileCheck %s --check-prefixes=CHECK,SIZE_LATE,SSE-SIZE_LATE,SSE2-SIZE_LATE
 ;
 ; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx  -cost-kind=latency      < %s | FileCheck %s --check-prefixes=CHECK,LATE,AVX-LATE,AVX1-LATE
-; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx  -cost-kind=code-size    < %s | FileCheck %s --check-prefixes=CHECK,SIZE,AVX-SIZE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx  -cost-kind=code-size    < %s | FileCheck %s --check-prefixes=CHECK,SIZE,AVX-SIZE,AVX1-SIZE
 ; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx  -cost-kind=size-latency < %s | FileCheck %s --check-prefixes=CHECK,SIZE_LATE,AVX-SIZE_LATE,AVX1-SIZE_LATE
 ;
 ; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx2 -cost-kind=latency      < %s | FileCheck %s --check-prefixes=CHECK,LATE,AVX-LATE,AVX2-LATE
-; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx2 -cost-kind=code-size    < %s | FileCheck %s --check-prefixes=CHECK,SIZE,AVX-SIZE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx2 -cost-kind=code-size    < %s | FileCheck %s --check-prefixes=CHECK,SIZE,AVX-SIZE,AVX2-SIZE
 ; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx2 -cost-kind=size-latency < %s | FileCheck %s --check-prefixes=CHECK,SIZE_LATE,AVX-SIZE_LATE,AVX2-SIZE_LATE
+;
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512f,+avx512vl -cost-kind=latency      < %s | FileCheck %s --check-prefixes=CHECK,LATE,AVX512-LATE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512f,+avx512vl -cost-kind=code-size    < %s | FileCheck %s --check-prefixes=CHECK,SIZE,AVX512-SIZE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512f,+avx512vl -cost-kind=size-latency < %s | FileCheck %s --check-prefixes=CHECK,SIZE_LATE,AVX512-SIZE_LATE
+;
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512bw,+avx512vl -cost-kind=latency      < %s | FileCheck %s --check-prefixes=CHECK,LATE,AVX512-LATE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512bw,+avx512vl -cost-kind=code-size    < %s | FileCheck %s --check-prefixes=CHECK,SIZE,AVX512-SIZE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512bw,+avx512vl -cost-kind=size-latency < %s | FileCheck %s --check-prefixes=CHECK,SIZE_LATE,AVX512-SIZE_LATE
+;
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mcpu=slm -cost-kind=latency      < %s | FileCheck %s --check-prefixes=CHECK,LATE,SSE-LATE,SLM-LATE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mcpu=slm -cost-kind=code-size    < %s | FileCheck %s --check-prefixes=CHECK,SIZE,SSE-SIZE,SLM-SIZE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mcpu=slm -cost-kind=size-latency < %s | FileCheck %s --check-prefixes=CHECK,SIZE_LATE,SSE-SIZE_LATE,SLM-SIZE_LATE
+;
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mcpu=goldmont -cost-kind=latency      < %s | FileCheck %s --check-prefixes=CHECK,LATE,SSE-LATE,GLM-LATE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mcpu=goldmont -cost-kind=code-size    < %s | FileCheck %s --check-prefixes=CHECK,SIZE,SSE-SIZE,GLM-SIZE
+; RUN: opt -mtriple=x86_64-- -passes="print<cost-model>" 2>&1 -disable-output -mcpu=goldmont -cost-kind=size-latency < %s | FileCheck %s --check-prefixes=CHECK,SIZE_LATE,SSE-SIZE_LATE,GLM-SIZE_LATE
 
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
 target triple = "x86_64-apple-macosx10.8.0"
@@ -300,38 +316,38 @@ define i32 @frem(i32 %arg) {
 }
 
 define i32 @fsqrt(i32 %arg) {
-; SSE-LATE-LABEL: 'fsqrt'
-; SSE-LATE-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
-; SSE-LATE-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
-; SSE-LATE-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
-; SSE-LATE-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
-; SSE-LATE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
-; SSE-LATE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
-; SSE-LATE-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
-; SSE-LATE-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
-; SSE-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
-;
-; SSE-SIZE-LABEL: 'fsqrt'
-; SSE-SIZE-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
-; SSE-SIZE-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
-; SSE-SIZE-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
-; SSE-SIZE-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
-; SSE-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
-; SSE-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
-; SSE-SIZE-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
-; SSE-SIZE-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
-; SSE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
-;
-; SSE-SIZE_LATE-LABEL: 'fsqrt'
-; SSE-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
-; SSE-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
-; SSE-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
-; SSE-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
-; SSE-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
-; SSE-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
-; SSE-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
-; SSE-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
-; SSE-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+; SSE2-LATE-LABEL: 'fsqrt'
+; SSE2-LATE-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
+; SSE2-LATE-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
+; SSE2-LATE-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
+; SSE2-LATE-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
+; SSE2-LATE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
+; SSE2-LATE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
+; SSE2-LATE-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
+; SSE2-LATE-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
+; SSE2-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; SSE2-SIZE-LABEL: 'fsqrt'
+; SSE2-SIZE-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
+; SSE2-SIZE-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
+; SSE2-SIZE-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
+; SSE2-SIZE-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
+; SSE2-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
+; SSE2-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
+; SSE2-SIZE-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
+; SSE2-SIZE-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
+; SSE2-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; SSE2-SIZE_LATE-LABEL: 'fsqrt'
+; SSE2-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
+; SSE2-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
+; SSE2-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
+; SSE2-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
+; SSE2-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
+; SSE2-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
+; SSE2-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
+; SSE2-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
+; SSE2-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
 ; AVX1-LATE-LABEL: 'fsqrt'
 ; AVX1-LATE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
@@ -344,6 +360,17 @@ define i32 @fsqrt(i32 %arg) {
 ; AVX1-LATE-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
 ; AVX1-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
+; AVX1-SIZE-LABEL: 'fsqrt'
+; AVX1-SIZE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
+; AVX1-SIZE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
+; AVX1-SIZE-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
+; AVX1-SIZE-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
+; AVX1-SIZE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
+; AVX1-SIZE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
+; AVX1-SIZE-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
+; AVX1-SIZE-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
+; AVX1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
 ; AVX1-SIZE_LATE-LABEL: 'fsqrt'
 ; AVX1-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
 ; AVX1-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
@@ -366,6 +393,17 @@ define i32 @fsqrt(i32 %arg) {
 ; AVX2-LATE-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
 ; AVX2-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
+; AVX2-SIZE-LABEL: 'fsqrt'
+; AVX2-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
+; AVX2-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
+; AVX2-SIZE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
+; AVX2-SIZE-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
+; AVX2-SIZE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
+; AVX2-SIZE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
+; AVX2-SIZE-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
+; AVX2-SIZE-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
+; AVX2-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
 ; AVX2-SIZE_LATE-LABEL: 'fsqrt'
 ; AVX2-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
 ; AVX2-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
@@ -377,6 +415,105 @@ define i32 @fsqrt(i32 %arg) {
 ; AVX2-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
 ; AVX2-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
+; AVX512-LATE-LABEL: 'fsqrt'
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX512-SIZE-LABEL: 'fsqrt'
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX512-SIZE_LATE-LABEL: 'fsqrt'
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; SLM-LATE-LABEL: 'fsqrt'
+; SLM-LATE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
+; SLM-LATE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
+; SLM-LATE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
+; SLM-LATE-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
+; SLM-LATE-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
+; SLM-LATE-NEXT:  Cost Model: Found an estimated cost of 70 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
+; SLM-LATE-NEXT:  Cost Model: Found an estimated cost of 140 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
+; SLM-LATE-NEXT:  Cost Model: Found an estimated cost of 280 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
+; SLM-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; SLM-SIZE-LABEL: 'fsqrt'
+; SLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
+; SLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
+; SLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
+; SLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
+; SLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
+; SLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 70 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
+; SLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 140 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
+; SLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 280 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
+; SLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; SLM-SIZE_LATE-LABEL: 'fsqrt'
+; SLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
+; SLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
+; SLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
+; SLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
+; SLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
+; SLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 70 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
+; SLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 140 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
+; SLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 280 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
+; SLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; GLM-LATE-LABEL: 'fsqrt'
+; GLM-LATE-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
+; GLM-LATE-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
+; GLM-LATE-NEXT:  Cost Model: Found an estimated cost of 74 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
+; GLM-LATE-NEXT:  Cost Model: Found an estimated cost of 148 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
+; GLM-LATE-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
+; GLM-LATE-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
+; GLM-LATE-NEXT:  Cost Model: Found an estimated cost of 134 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
+; GLM-LATE-NEXT:  Cost Model: Found an estimated cost of 268 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
+; GLM-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; GLM-SIZE-LABEL: 'fsqrt'
+; GLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
+; GLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
+; GLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 74 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
+; GLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 148 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
+; GLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
+; GLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
+; GLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 134 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
+; GLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 268 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
+; GLM-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; GLM-SIZE_LATE-LABEL: 'fsqrt'
+; GLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %F32 = call float @llvm.sqrt.f32(float undef)
+; GLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
+; GLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 74 for instruction: %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
+; GLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 148 for instruction: %V16F32 = call <16 x float> @llvm.sqrt.v16f32(<16 x float> undef)
+; GLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %F64 = call double @llvm.sqrt.f64(double undef)
+; GLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %V2F64 = call <2 x double> @llvm.sqrt.v2f64(<2 x double> undef)
+; GLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 134 for instruction: %V4F64 = call <4 x double> @llvm.sqrt.v4f64(<4 x double> undef)
+; GLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 268 for instruction: %V8F64 = call <8 x double> @llvm.sqrt.v8f64(<8 x double> undef)
+; GLM-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
   %F32 = call float @llvm.sqrt.f32(float undef)
   %V4F32 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> undef)
   %V8F32 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> undef)
@@ -457,6 +594,39 @@ define i32 @fabs(i32 %arg) {
 ; AVX-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = call <8 x double> @llvm.fabs.v8f64(<8 x double> undef)
 ; AVX-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
+; AVX512-LATE-LABEL: 'fabs'
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %F32 = call float @llvm.fabs.f32(float undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = call <4 x float> @llvm.fabs.v4f32(<4 x float> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = call <8 x float> @llvm.fabs.v8f32(<8 x float> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = call <16 x float> @llvm.fabs.v16f32(<16 x float> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %F64 = call double @llvm.fabs.f64(double undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2F64 = call <2 x double> @llvm.fabs.v2f64(<2 x double> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = call <4 x double> @llvm.fabs.v4f64(<4 x double> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = call <8 x double> @llvm.fabs.v8f64(<8 x double> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX512-SIZE-LABEL: 'fabs'
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %F32 = call float @llvm.fabs.f32(float undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = call <4 x float> @llvm.fabs.v4f32(<4 x float> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = call <8 x float> @llvm.fabs.v8f32(<8 x float> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = call <16 x float> @llvm.fabs.v16f32(<16 x float> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %F64 = call double @llvm.fabs.f64(double undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2F64 = call <2 x double> @llvm.fabs.v2f64(<2 x double> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = call <4 x double> @llvm.fabs.v4f64(<4 x double> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = call <8 x double> @llvm.fabs.v8f64(<8 x double> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX512-SIZE_LATE-LABEL: 'fabs'
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %F32 = call float @llvm.fabs.f32(float undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = call <4 x float> @llvm.fabs.v4f32(<4 x float> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = call <8 x float> @llvm.fabs.v8f32(<8 x float> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = call <16 x float> @llvm.fabs.v16f32(<16 x float> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %F64 = call double @llvm.fabs.f64(double undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2F64 = call <2 x double> @llvm.fabs.v2f64(<2 x double> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = call <4 x double> @llvm.fabs.v4f64(<4 x double> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = call <8 x double> @llvm.fabs.v8f64(<8 x double> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
   %F32 = call float @llvm.fabs.f32(float undef)
   %V4F32 = call <4 x float> @llvm.fabs.v4f32(<4 x float> undef)
   %V8F32 = call <8 x float> @llvm.fabs.v8f32(<8 x float> undef)
@@ -537,6 +707,39 @@ define i32 @fcopysign(i32 %arg) {
 ; AVX-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = call <8 x double> @llvm.copysign.v8f64(<8 x double> undef, <8 x double> undef)
 ; AVX-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
+; AVX512-LATE-LABEL: 'fcopysign'
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %F32 = call float @llvm.copysign.f32(float undef, float undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = call <4 x float> @llvm.copysign.v4f32(<4 x float> undef, <4 x float> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = call <8 x float> @llvm.copysign.v8f32(<8 x float> undef, <8 x float> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = call <16 x float> @llvm.copysign.v16f32(<16 x float> undef, <16 x float> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %F64 = call double @llvm.copysign.f64(double undef, double undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2F64 = call <2 x double> @llvm.copysign.v2f64(<2 x double> undef, <2 x double> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = call <4 x double> @llvm.copysign.v4f64(<4 x double> undef, <4 x double> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = call <8 x double> @llvm.copysign.v8f64(<8 x double> undef, <8 x double> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX512-SIZE-LABEL: 'fcopysign'
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %F32 = call float @llvm.copysign.f32(float undef, float undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = call <4 x float> @llvm.copysign.v4f32(<4 x float> undef, <4 x float> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = call <8 x float> @llvm.copysign.v8f32(<8 x float> undef, <8 x float> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = call <16 x float> @llvm.copysign.v16f32(<16 x float> undef, <16 x float> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %F64 = call double @llvm.copysign.f64(double undef, double undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2F64 = call <2 x double> @llvm.copysign.v2f64(<2 x double> undef, <2 x double> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = call <4 x double> @llvm.copysign.v4f64(<4 x double> undef, <4 x double> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = call <8 x double> @llvm.copysign.v8f64(<8 x double> undef, <8 x double> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX512-SIZE_LATE-LABEL: 'fcopysign'
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %F32 = call float @llvm.copysign.f32(float undef, float undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = call <4 x float> @llvm.copysign.v4f32(<4 x float> undef, <4 x float> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = call <8 x float> @llvm.copysign.v8f32(<8 x float> undef, <8 x float> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = call <16 x float> @llvm.copysign.v16f32(<16 x float> undef, <16 x float> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %F64 = call double @llvm.copysign.f64(double undef, double undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2F64 = call <2 x double> @llvm.copysign.v2f64(<2 x double> undef, <2 x double> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = call <4 x double> @llvm.copysign.v4f64(<4 x double> undef, <4 x double> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = call <8 x double> @llvm.copysign.v8f64(<8 x double> undef, <8 x double> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
   %F32 = call float @llvm.copysign.f32(float undef, float undef)
   %V4F32 = call <4 x float> @llvm.copysign.v4f32(<4 x float> undef, <4 x float> undef)
   %V8F32 = call <8 x float> @llvm.copysign.v8f32(<8 x float> undef, <8 x float> undef)
@@ -617,6 +820,39 @@ define i32 @fma(i32 %arg) {
 ; AVX-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %V8F64 = call <8 x double> @llvm.fma.v8f64(<8 x double> undef, <8 x double> undef, <8 x double> undef)
 ; AVX-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
+; AVX512-LATE-LABEL: 'fma'
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = call float @llvm.fma.f32(float undef, float undef, float undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = call <4 x float> @llvm.fma.v4f32(<4 x float> undef, <4 x float> undef, <4 x float> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = call <8 x float> @llvm.fma.v8f32(<8 x float> undef, <8 x float> undef, <8 x float> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = call <16 x float> @llvm.fma.v16f32(<16 x float> undef, <16 x float> undef, <16 x float> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = call double @llvm.fma.f64(double undef, double undef, double undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = call <2 x double> @llvm.fma.v2f64(<2 x double> undef, <2 x double> undef, <2 x double> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = call <4 x double> @llvm.fma.v4f64(<4 x double> undef, <4 x double> undef, <4 x double> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = call <8 x double> @llvm.fma.v8f64(<8 x double> undef, <8 x double> undef, <8 x double> undef)
+; AVX512-LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX512-SIZE-LABEL: 'fma'
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = call float @llvm.fma.f32(float undef, float undef, float undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = call <4 x float> @llvm.fma.v4f32(<4 x float> undef, <4 x float> undef, <4 x float> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = call <8 x float> @llvm.fma.v8f32(<8 x float> undef, <8 x float> undef, <8 x float> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = call <16 x float> @llvm.fma.v16f32(<16 x float> undef, <16 x float> undef, <16 x float> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = call double @llvm.fma.f64(double undef, double undef, double undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = call <2 x double> @llvm.fma.v2f64(<2 x double> undef, <2 x double> undef, <2 x double> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = call <4 x double> @llvm.fma.v4f64(<4 x double> undef, <4 x double> undef, <4 x double> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = call <8 x double> @llvm.fma.v8f64(<8 x double> undef, <8 x double> undef, <8 x double> undef)
+; AVX512-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX512-SIZE_LATE-LABEL: 'fma'
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = call float @llvm.fma.f32(float undef, float undef, float undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = call <4 x float> @llvm.fma.v4f32(<4 x float> undef, <4 x float> undef, <4 x float> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = call <8 x float> @llvm.fma.v8f32(<8 x float> undef, <8 x float> undef, <8 x float> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = call <16 x float> @llvm.fma.v16f32(<16 x float> undef, <16 x float> undef, <16 x float> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = call double @llvm.fma.f64(double undef, double undef, double undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = call <2 x double> @llvm.fma.v2f64(<2 x double> undef, <2 x double> undef, <2 x double> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = call <4 x double> @llvm.fma.v4f64(<4 x double> undef, <4 x double> undef, <4 x double> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = call <8 x double> @llvm.fma.v8f64(<8 x double> undef, <8 x double> undef, <8 x double> undef)
+; AVX512-SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
   %F32 = call float @llvm.fma.f32(float undef, float undef, float undef)
   %V4F32 = call <4 x float> @llvm.fma.v4f32(<4 x float> undef, <4 x float> undef, <4 x float> undef)
   %V8F32 = call <8 x float> @llvm.fma.v8f32(<8 x float> undef, <8 x float> undef, <8 x float> undef)